Advanced vacant lot detection systems are becoming an integral part of modern outdoor parking fields. As most of the parking fields are equipped with CCTV surveillance, vision-based solution has emerged as an effective, low-cost alternative. Although, image processing-based solutions suffer from issues due to irregular illumination, climatic variations, partial occlusion and perspective distortion due to camera placement, deep learning-based approaches have proved to be resilient. Another typical problem is the transient occlusion of the lots from the camera view due to passing-by vehicles and people. This paper presents a robust method for vacant lot detection by combining merits of deep learning approach with motion tracking to overcome the above issues. The fundamental idea is to find out a static frame which is free of any moving objects and submit it to a deep learning model for counting and localisation of vacant lots. Assuming smooth and streamlined motion of objects, motion tracking is implemented with constant velocity Kalman filter. Presence of motion is decided by checking alive-tracks in the current frame. Once a motion-free frame is recognised, the parking lot portion in the frame is cropped and input to a pre-trained YOLO (You Only Look Once) CNN (Convolutional Neural Network) model for counting and localisation of cars. Number of vacant lots are found by subtracting this count from the total number of available parking lots and their locations are determined by cross-matching detected bounding boxes of the cars against the pre-calculated locations of all the parking lots. © 2018, Institute of Advanced Scientific Research, Inc. All rights reserved.
cited By 0
E. K. Jose and S. Veni, “YOLO classification with multiple object tracking for vacant parking lot detection”, Journal of Advanced Research in Dynamical and Control Systems, vol. 10, pp. 683-689, 2018.